Sonar Image Target Detection Based on Style Transfer Learning and Random Shape of Noise under Zero Shot Target

With the development of sonar technology, images have been widely used to detect targets. However, there are many challenges for in terms object detection. For example, detectable targets data more sparse than those optical images, real underwater scanning experiment is complicated, and image styles produced by different types equipment due their characteristics inconsistent, which makes it difficult use them detection recognition algorithms. In order solve these problems, we propose a novel object-detection method based on style learning random noise with various shapes. Sonar target sample generated through transfer, enhances insufficient objects image. By introducing shapes, included points, lines, rectangles, problems mud sand obstruction mutilated environment solved, single poses improved fusing multiple target. meantime, feature enhancement proposed issue missing key features when using transfer directly. The experimental results show that our achieves better precision.